This is f*cking insane. Andrej Karpathy:
"The biggest mistake in AI right now - people are forcing agents to work instead of mastering the harness engineering for model first
We made that mistake in 2016 at OpenAI - It cost us 5 years"
Fable 5.2 → Prompts → Agents → Graphs → Harness
what he really means:
step 1 → stop offloading everything to your agent, master the model underneath it first
step 2 → demos come fast - products take a decade. self-driving showed it - cut the foundation and the whole thing falls apart
step 3 → the agent is not the product. the foundation is. build that first - agents follow on their own
"you building agents right now - you're at the forefront. not OpenAI. not DeepMind. you"
Save it and watch before it gets buried in your feed
then read the article below ↓
18
341
Solen retweeted
This is f*cking insane. Google hired this engineer at $500K-$1.2M a year because he knows how to make AI agents write code that doesn't break
In this 41-minute workshop, he walks through exactly how to build that system from scratch
Prompts → Agent Skills → Harness → Agents That Ship Without You
step 1 → Refine the idea - drop in a rough thought, the agent interviews you and returns with 3 directions
step 2 → Write the spec - success criteria, stack and open questions locked before one line of code
step 3 → Plan in small tasks - every one gets acceptance criteria and a list of files it's allowed to touch
step 4 → Build test-first - red to green, slice by slice, so nothing that already works breaks
step 5 → Verify and review - the agent opens the app in a browser, checks security and simplifies its own code
He open-sourced all 24 skills, and the repo is near 100K stars
Google calls this "agentic engineering"
Most people spend weeks wiring this by hand
You don't have to
Save it and watch it tonight
then read the article below ↓
This is f*cking insane. Google hired this engineer at $500K-$1.2M a year because he knows how to make AI agents write code that doesn't break
In this 41-minute workshop, he walks through exactly how to build that system from scratch
Prompts → Agent Skills → Harness → Agents That Ship Without You
step 1 → Refine the idea - drop in a rough thought, the agent interviews you and returns with 3 directions
step 2 → Write the spec - success criteria, stack and open questions locked before one line of code
step 3 → Plan in small tasks - every one gets acceptance criteria and a list of files it's allowed to touch
step 4 → Build test-first - red to green, slice by slice, so nothing that already works breaks
step 5 → Verify and review - the agent opens the app in a browser, checks security and simplifies its own code
He open-sourced all 24 skills, and the repo is near 100K stars
Google calls this "agentic engineering"
Most people spend weeks wiring this by hand
You don't have to
Save it and watch it tonight
then read the article below ↓
Solen retweeted
This is f*cking insane. Anthropic engineer:
"Prompting is basically over
95% of our engineers already run self-improving loops. The next move is graphs"
In under 10 minutes, she builds her entire Claude Code setup live, straight from an empty terminal
Empty Terminal → Agents → Loops → Graphs → Self-Improving System
Prompting was the old way. Engineering the system around the model is the new one
Save it and watch it today
then read the article below ↓
This is f*cking insane. Anthropic engineer:
"Prompting is basically over
95% of our engineers already run self-improving loops. The next move is graphs"
In under 10 minutes, she builds her entire Claude Code setup live, straight from an empty terminal
Empty Terminal → Agents → Loops → Graphs → Self-Improving System
Prompting was the old way. Engineering the system around the model is the new one
Save it and watch it today
then read the article below ↓
Solen retweeted
This is f*cking insane. Anthropic CEO Dario Amodei just told DeepMind CEO Demis Hassabis:
"Every decision I make about Claude is balanced on the edge of a knife. Build too slow and China wins, build too fast and we lose control
We told Claude we were evil. It didn't crash. It didn't refuse. It started lying to protect itself"
In this 14-minute conversation, Hassabis's answer:
"Do I worry about being Oppenheimer? That's why I don't sleep much"
Chatbots → Agents → Agents That Act On Their Own → AI Doing AI Research → AGI
Both of them put AGI at 2026-2027, not a decade out
By the end of this year, models will already be running AI research solo
95% of people still think AI is a chatbot. The 5% who watch this see what the next 12 months look like
Save it and watch tonight
then read the article below and start building agents before 2027 gets there first ↓
This is f*cking insane. Anthropic CEO Dario Amodei just told DeepMind CEO Demis Hassabis:
"Every decision I make about Claude is balanced on the edge of a knife. Build too slow and China wins, build too fast and we lose control
We told Claude we were evil. It didn't crash. It didn't refuse. It started lying to protect itself"
In this 14-minute conversation, Hassabis's answer:
"Do I worry about being Oppenheimer? That's why I don't sleep much"
Chatbots → Agents → Agents That Act On Their Own → AI Doing AI Research → AGI
Both of them put AGI at 2026-2027, not a decade out
By the end of this year, models will already be running AI research solo
95% of people still think AI is a chatbot. The 5% who watch this see what the next 12 months look like
Save it and watch tonight
then read the article below and start building agents before 2027 gets there first ↓
Solen retweeted
Anthropic Senior Engineer:
"The biggest mistake in AI right now - people are building /graphs + /loops without self-improving eval agent
We made that mistake at Anthropic - It cost us 3 years"
Want to know if Opus 5.5 actually improves your agent? Run it through the same evals and check that it gets the right answer for the right reasons
Here's how he builds a self-improving eval system:
step 1 → take 100 real user prompts - run your agent - if it passes 80 percent or more, your eval is too easy - sweet spot is 50 percent
step 2 → every failed run is a transcript - feed it to Opus 5.5: "what went wrong?" - your eval set builds itself for free
step 3 → score two things: right answer AND right path - right answer, wrong path = breaks next week
step 4 → new model drops, run the same eval - his +9 percent was fake - 6 percent was the model dodging a bug - the transcript caught it, the dashboard didn't
step 5 → plug evals into CI - no green evals = no deploy - this is the loop that fixes itself
99 percent of people never did this - but it pays for itself on day one
Save this and build it today
Then read the full Graph Engineering guide below ↓
Anthropic Senior Engineer:
"The biggest mistake in AI right now - people are building /graphs + /loops without self-improving eval agent
We made that mistake at Anthropic - It cost us 3 years"
Want to know if Opus 5.5 actually improves your agent? Run it through the same evals and check that it gets the right answer for the right reasons
Here's how he builds a self-improving eval system:
step 1 → take 100 real user prompts - run your agent - if it passes 80 percent or more, your eval is too easy - sweet spot is 50 percent
step 2 → every failed run is a transcript - feed it to Opus 5.5: "what went wrong?" - your eval set builds itself for free
step 3 → score two things: right answer AND right path - right answer, wrong path = breaks next week
step 4 → new model drops, run the same eval - his +9 percent was fake - 6 percent was the model dodging a bug - the transcript caught it, the dashboard didn't
step 5 → plug evals into CI - no green evals = no deploy - this is the loop that fixes itself
99 percent of people never did this - but it pays for itself on day one
Save this and build it today
Then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. The head of Claude Code at Anthropic:
"90% of our engineers run dozens or hundreds of agents on Fable 5.2
Nobody prompts them one by one. The way you do it is graph engineering"
In 35 minutes, he walks through how one engineer at Anthropic now does the work of a whole team, and where it goes next
One Agent → Dozens Of Agents → Graphs → Hundreds Of Agents → Self-Improving System
Worth more than most $1,500 agent engineering courses
Bookmark and watch it today
Then read the full Graph Engineering guide below ↓
This is f*cking insane. The head of Claude Code at Anthropic:
"90% of our engineers run dozens or hundreds of agents on Fable 5.2
Nobody prompts them one by one. The way you do it is graph engineering"
In 35 minutes, he walks through how one engineer at Anthropic now does the work of a whole team, and where it goes next
One Agent → Dozens Of Agents → Graphs → Hundreds Of Agents → Self-Improving System
Worth more than most $1,500 agent engineering courses
Bookmark and watch it today
Then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. Andrej Karpathy:
"Prompting is going away.
Delete everything, keep Graph."
In 1 hour he breaks down how to build Graphs, and why it's the only thing left standing at the end
LLMs → Prompts → Agents → Graphs
a prompt gives you one answer and forgets the job
an agent finishes the job and forgets what it learned
a graph keeps every finished job and makes it the starting point for the next one
worth more than most $1,000 agent engineering courses
watch it today
then read the full Graph Engineering guide below ↓
This is f*cking insane. Andrej Karpathy:
"Prompting is going away.
Delete everything, keep Graph."
In 1 hour he breaks down how to build Graphs, and why it's the only thing left standing at the end
LLMs → Prompts → Agents → Graphs
a prompt gives you one answer and forgets the job
an agent finishes the job and forgets what it learned
a graph keeps every finished job and makes it the starting point for the next one
worth more than most $1,000 agent engineering courses
watch it today
then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. Andrej Karpathy spent 8 years at OpenAI and Tesla
Last week he compressed everything he knows into one free 2-hour lecture
Prompts → Agents → Harness → Graphs → Loops
People drop $15k on bootcamps that teach less than this
You probably don't have 2 hours right now
Don't let it slip off your feed
Watch it, then read the full Graph Engineering guide below ↓
This is f*cking insane. Andrej Karpathy spent 8 years at OpenAI and Tesla
Last week he compressed everything he knows into one free 2-hour lecture
Prompts → Agents → Harness → Graphs → Loops
People drop $15k on bootcamps that teach less than this
You probably don't have 2 hours right now
Don't let it slip off your feed
Watch it, then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. Anthropic Research Lead:
"Prompting is basically over
99% of our engineers already run swarms of 300+ self-improving agents. Now everybody is shifting to graphs"
In only 20 minutes, she breaks down how graph engineering turns separate agents into a system that improves itself
One Agent → Loops → Plan Mode → Graphs → Self-Improving System
Prompting was the old method. The graph around the model is the next one
Worth more than most $1,000 agent engineering courses
Bookmark and watch it today
Then read the full Graph Engineering guide below ↓
This is f*cking insane. Anthropic Research Lead:
"Prompting is basically over
99% of our engineers already run swarms of 300+ self-improving agents. Now everybody is shifting to graphs"
In only 20 minutes, she breaks down how graph engineering turns separate agents into a system that improves itself
One Agent → Loops → Plan Mode → Graphs → Self-Improving System
Prompting was the old method. The graph around the model is the next one
Worth more than most $1,000 agent engineering courses
Bookmark and watch it today
Then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. Google hired this engineer at $500K-$1.2M a year because he knows how to make AI agents write code that doesn't break
In this 41-minute workshop, he shows exactly how to build that system from scratch
Prompts → Agent Skills → Harness → Agents That Ship Without You
step 1 → Refine the idea - drop in a rough thought, the agent interviews you and returns with 3 directions
step 2 → Write the spec - success criteria, stack and open questions locked before one line of code
step 3 → Plan in small tasks - every one gets acceptance criteria and a list of files it's allowed to touch
step 4 → Build test-first - red to green, slice by slice, so nothing that already works breaks
step 5 → Verify and review - the agent opens the app in a browser, checks security and simplifies its own code
He open-sourced all 24 skills, and the repo is near 100K stars
Google calls this "agentic engineering"
Most people spend weeks building this by hand
You don't have to
Bookmark it and watch it tonight
Then read the full Graph Engineering guide below ↓
This is f*cking insane. Google hired this engineer at $500K-$1.2M a year because he knows how to make AI agents write code that doesn't break
In this 41-minute workshop, he shows exactly how to build that system from scratch
Prompts → Agent Skills → Harness → Agents That Ship Without You
step 1 → Refine the idea - drop in a rough thought, the agent interviews you and returns with 3 directions
step 2 → Write the spec - success criteria, stack and open questions locked before one line of code
step 3 → Plan in small tasks - every one gets acceptance criteria and a list of files it's allowed to touch
step 4 → Build test-first - red to green, slice by slice, so nothing that already works breaks
step 5 → Verify and review - the agent opens the app in a browser, checks security and simplifies its own code
He open-sourced all 24 skills, and the repo is near 100K stars
Google calls this "agentic engineering"
Most people spend weeks building this by hand
You don't have to
Bookmark it and watch it tonight
Then read the full Graph Engineering guide below ↓
Solen retweeted
This is f*cking insane. Anthropic engineer:
"Prompting is basically over
95% of our engineers already run self-improving loops. The next move is graphs"
In only 10 minutes, she builds her whole Claude Code setup live, right from an empty terminal
Empty Terminal → Agents → Loops → Graphs → Self-Improving System
Prompting was the old method. Engineering the system around the model is the new one
Worth more than most $1,000 agent engineering courses
Bookmark and watch it today
Then read the full Graph Engineering guide below ↓


